Generating unique IDs for multiple sources

Question

Consider N sources of data, each with a stream of events

Event{
    long id;
    Object data;
}

Some of the events within one stream might have the same id, as events might span across Updated, New etc. So we can see the following two streams:

<1, 2, 3, 1, 5, 2>
<3, 3, 4, 5, 4>

I would now like to combine these into one stream st each order id is definitely going to be unique.

The easy way would be to use a String instead of long and append source number, generating sth like:

<"1 - 1", "1 - 2", "1 - 3", "2-3", "2-3" ... >

Is there a more memory coimpact way/better approach?

Answer 1

Your String solution is fine and in fact quite common. If you're interested in making it more compact, you may want to use a tuple of integers.

Another common method used in distributed systems is to use range allocation: have a central (singleton) server which allocates ranges in which each client can name its IDs. Such server could allocate, for example, the range 0-99 to client1, 100-199 to client2 etc. When a client exhausts the range it was allocated, it contacts the server again to allocate a new range.

Answer 2

Depending on the ranges of your stream/event numbers, you could combine the two numbers into a single int or long, placing the stream number in the top so many bits and the event number in the bottom so many bits. For example:

public static int getCombinedNo(int streamNo, int eventNo) {
    if (streamNo >= (1 << 16))
      throw new IllegalArgumentException("Stream no too big");
    if (eventNo >= (1 << 16))
      throw new IllegalArgumentException("Event no too big");
    return (streamNo << 16) | eventNo;
}

This will only use 4 bytes per int as opposed to in the order of (say) 50-ish bytes for a typical String of the type you mention. (In this case, it also assumes that neither stream nor event number will exceed 65535.)

But: your string solution is also nice and clear. Is memory really that tight that you can't spare an extra 50 bytes per event?